Smarter Pest Identification Technology (SPIDTECH) : digital identification of insect pest and crop disease using convolutional neural networks

Correct insect pest and crop disease identification plays a vital role in crop protection and management as it serves as a factor in determining suitable pest and disease control methods and management practices in farms. This study focuses on building an Android application named Smarter Pest Identification Technology (SPIDTECH) that aims to guide its users through digital identification of insect pest and disease of rice, corn, coffee, cacao, banana, sugarcane, coconut, soybean, and tomato in the Philippines. The system uses Inception-v3, a pre-trained image recognition model, comprised of a feature-extraction layer with convolutional neural network and a classification layer with a fully connected and softmax layer. The model was pre-trained from ImageNet dataset and can classify 1000 object categories. The model was retained, in a progress called transfer learning, with insect pest and disease dataset comprised of no less than 500 images of each pest and disease of each crop captured from different sites in the Philippines. The transfer learning process utilizes the feature extraction capability of the pre-trained model to classify input images using new dataset. This study uses a total of 18 models: two models for each of the nine focus crops, one for the pest and one for the disease identification. The model accepts image input and outputs a ranked classification based on accuracy. The model extracts general features from the input image and classifies them based on those features. Along with the identification results, the application also includes a library data comprised of relevant information about each pest and disease such as identification signs, life cycle, and management practice to help its users in correctly managing the pest and disease. Moreover, the application logs data from its users such as GPS location, top results, and image data for tracking and remote monitoring of pest and disease occurrence.